Name can be indicative of similarity if we can extract the right words
Many descriptions are NOT useful for similarity, as they're actually about the context of the object, not the object itself
Dimensions are not very useful -- a big statue and a small statue are more similar than a big painting and a big statue.
Materials are useful, although having a stronger hierarchy would improve this
Classifications are like the most useful
Artists' styles can vary dramatically, worth including though
Date is not a requirement (like it is for person) as a 20th C copy of a 12th Century painting are clearly more similar than two 20th C works, but could be useful
But the similarity should also take into account similarity of the carried/shown work(s).
To wit, a book about a Torosaurus and a poster about a Torosaurus are (I would say) more similar than a book about a Torosaurus and a book about particle physics. However the materiality (e.g. the properties of the object) is clearly the same -- they're both books.
To generate such a query we would need to retrieve the works and process them. Which for objects that carry many works could be a challenge if we want to limit the clauses in the query for performance reasons. E.g. we can't just do
Objects are interesting in terms of similarity.
But the similarity should also take into account similarity of the carried/shown work(s).
To wit, a book about a Torosaurus and a poster about a Torosaurus are (I would say) more similar than a book about a Torosaurus and a book about particle physics. However the materiality (e.g. the properties of the object) is clearly the same -- they're both books.
To generate such a query we would need to retrieve the works and process them. Which for objects that carry many works could be a challenge if we want to limit the clauses in the query for performance reasons. E.g. we can't just do
in the object's similar() query as n might be high, and the cost of the embedded similar()s might also be high.